Bryan Liles is a senior principal engineer at Amazon Web Services, working with the AWS Data and S3 organization.
He addresses complex technical challenges to improve performance across AWS Analytics and S3 services while mentoring engineers.
With over 30 years of experience in enterprise technologies and cloud infrastructure, Liles provides significant expertise.
During the recent AI NEXTGEN America Conference, Liles gave a keynote address on the importance of building things in artificial intelligence and how models are built.
Liles said that, in artificial intelligence, weighted inputs are the fundamental building block of a neuron.
Each input (X1, X2, X3) is multiplied by a corresponding weight (W1, W2), and these weighted values are summed.
The bias enables the model to shift the decision boundary independently of the inputs. The result Z is the neuron's raw output before activation.
This linear combination is the core computation every neuron in a neural network performs. After this computation, the network passes the result through an activation function.
Without multiple layers, the network would only produce a linear transformation.
The activation function introduces non-linearity, enabling the network to learn complex patterns such as curves, edges, and abstract features.
Common activation functions include sigmoid and tanh. The network then passes the output A to the next layer, where it evaluates its prediction.
The loss function compares the network's prediction with the actual target value. For example, if the prediction is 0.7 and the target is 1.0, the squared error is 0.9. Squaring the error emphasizes larger discrepancies between the prediction and the target.
The primary objective of training is to minimize loss; lower loss values indicate better model performance.
Different loss functions suit different tasks, such as cross-entropy for classification problems. As model parameters are adjusted, researchers can build larger, more complex models.
Liles emphasizes the importance of engaging with real-world problems, whether manual or cognitive, and highlights the need for personal responsibility.
Artificial intelligence is not a universal solution. Individuals develop expertise through iterative processes of building, learning, receiving feedback, making adjustments, and persisting despite repeated failures.
However, mere exposure does not equate to experience.
Liles recommends actively building projects using freely available tools.
Information represents output, while capability reflects outcome.
Working on meaningful problems, getting actionable feedback, and repeatedly refining your approach increases the likelihood of achieving impactful results.
